Demand Estimation in Research-Based Forecasts

I have had the honor and privilege of teaching research methods, statistics, and economics at The University of Texas at Austin and St. Edward's University over the past several years. I have taught undergraduate and MBA students, all while maintaining my full-time consulting practice. In this post, I reflect on the synergies produced between these various endeavors, in particular focusing on forecasting as a common consulting engagement that draws heavily from all three areas.

While there are many elements that go into the development of a forecast, I will address demand estimation as an input to a research-based forecast model in this post. When I teach economics, we cover a variety of methods that include conducting surveys of consumers or other decision makers about future events or decisions and reviewing historical data.

Surveys can be used to test the market's likely response to future events. These experiments can be simple and direct: would you purchase X at $Y if Z happened? They can also be more subtle and complex, involving the establishment of baseline behavior followed by the presentation of one or more scenarios involving product launches, price changes, communication campaigns, and/or product/service specifications. These more complex survey designs often include experimental plans that support conjoint analysis or discrete choice modeling, wherein scenario dimensions and product/service characteristics are varied in accordance with principles of balance and orthogonality to support predictions for ranges of possibility not explicitly covered by the survey design. At the end of the research process, however, you have still basically asked the same question: would you purchase X at $Y if Z happened? And, thus, the answer, or space of answers, is based on hypothetical, prospective estimates of future behavior. Such results are generally viewed as optimistic estimates of steady-state, efficient market conditions that need to be tempered in order to calibrate a forecast to real-world dynamics that delay adoption while word of mouth spreads (or confirms) what is known about the product/service tested in the research setting. After all, research, by design, generally strives to fully engage respondents in the decision process that we are studying, a state that is far less controllable in the real-world marketplace.

Regression analysis and other related approaches can be used to produce demand curves that are functions of a variety of measureable market conditions. Examples include recent period sales of the same or similar goods or services, broader market and economic conditions, production costs, levels of promotion and selling, and price point, just to name a few. Such historical models can be used in conjunction with models or results from surveys to allow a forecast to be informed by the rate of adoption experienced by precedent products under various real-world market conditions.

Taken together, along with common sense business acumen, these research methods can produce sound predictions about the likelihood that an individual or organization will adopt or purchase a product or service. The final factor, at least that I will discuss in this post, needed to inform a research-based forecast is an estimate of the size of the addressable market. Here we also turn to research, usually drawing on a variety of secondary sources to establish the number of consumers, patients, or organizations that broadly define a market and any established relevant segmentation schemes. Additional primary research then explores and subsequently quantifies the various decision flows and process steps that might be required to bridge the gap between the broadly defined marketplace and one that more closely resembles the relevant real-world situation. As an example, census data can provide estimates of households by geography and income; other research reports can provide estimates of the prevalence of high speed Internet access cut along those same dimensions. Primary research can then establish the relative size of consumer segments that have been shown to require differentiated marketing communications strategies.